Unable to run resnet_34_8s_demo.ipynb

[sharanry]

TypeError: torch.FloatTensor constructor received an invalid combination of arguments - got (int, int, numpy.int64, numpy.int64), but expected one of:
 * no arguments
 * (int ...)
      didn't match because some of the arguments have invalid types: (int, int, numpy.int64, numpy.int64)
 * (torch.FloatTensor viewed_tensor)
 * (torch.Size size)
 * (torch.FloatStorage data)
 * (Sequence data)

Which version of pytorch is to be used?

[warmspringwinds]

@sharanry Hi,

It should work with the latest available version,

Could you, please, give me the line where exception occurred ?

[superoil1983]

---

TypeError Traceback (most recent call last)

in () **----> 1 fcn = resnet_dilated.Resnet34_8s(num_classes=21)** 2 fcn.load_state_dict(torch.load('resnet_34_8s_66.pth')) 3 fcn.cuda() 4 fcn.eval() ### related lines： **/pytorch-segmentation-detection/vision/torchvision/models/resnet.py** in __init__(self, inplanes, planes, stride, downsample, dilation) **---> 45** self.conv1 = conv3x3(inplanes, planes, stride, dilation=dilation) 46 self.bn1 = nn.BatchNorm2d(planes) **/pytorch-segmentation-detection/vision/torchvision/models/resnet.py** in conv3x3(in_planes, out_planes, stride, dilation) 36 return nn.Conv2d(in_planes, out_planes, kernel_size=kernel_size, stride=stride, **---> 37** padding=full_padding, dilation=dilation, bias=False) **/usr/local/lib/python3.5/dist-packages/torch/nn/modules/conv.py** in __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, transposed, output_padding, groups, bias) 32 self.weight = Parameter(torch.Tensor( **---> 33** out_channels, in_channels // groups, *kernel_size)) 34 if bias: TypeError: torch.FloatTensor constructor received an invalid combination of arguments - got (int, int, numpy.int64, numpy.int64), but expected one of: ...

Probably the same error.

[coolcoder001]

Probably the same error : fcn = resnet_dilated.Resnet18_8s(num_classes=2) throwing me error like this :

TypeError                                 Traceback (most recent call last)
<ipython-input-3-61af6330ec33> in <module>()
----> 1 fcn = resnet_dilated.Resnet18_8s(num_classes=2)

/home/xxxx/pytorch-segmentation-detection/pytorch_segmentation_detection/models/resnet_dilated.py in __init__(self, num_classes)
     54                                       pretrained=True,
     55                                       output_stride=8,
---> 56                                       remove_avg_pool_layer=True)
     57 
     58         # Randomly initialize the 1x1 Conv scoring layer

/home/xxxx/pytorch-segmentation-detection/vision/torchvision/models/resnet.py in resnet18(pretrained, **kwargs)
    224         pretrained (bool): If True, returns a model pre-trained on ImageNet
    225     """
--> 226     model = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs)
    227     if pretrained:
    228         model.load_state_dict(model_zoo.load_url(model_urls['resnet18']))

/home/xxxx/pytorch-segmentation-detection/vision/torchvision/models/resnet.py in __init__(self, block, layers, num_classes, fully_conv, remove_avg_pool_layer, output_stride)
    141         self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
    142 
--> 143         self.layer1 = self._make_layer(block, 64, layers[0])
    144         self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
    145         self.layer3 = self._make_layer(block, 256, layers[2], stride=2)

/home/xxxx/pytorch-segmentation-detection/vision/torchvision/models/resnet.py in _make_layer(self, block, planes, blocks, stride, dilation)
    189 
    190         layers = []
--> 191         layers.append(block(self.inplanes, planes, stride, downsample, dilation=self.current_dilation))
    192         self.inplanes = planes * block.expansion
    193         for i in range(1, blocks):

/home/xxxx/pytorch-segmentation-detection/vision/torchvision/models/resnet.py in __init__(self, inplanes, planes, stride, downsample, dilation)
     43     def __init__(self, inplanes, planes, stride=1, downsample=None, dilation=1):
     44         super(BasicBlock, self).__init__()
---> 45         self.conv1 = conv3x3(inplanes, planes, stride, dilation=dilation)
     46         self.bn1 = nn.BatchNorm2d(planes)
     47         self.relu = nn.ReLU(inplace=True)

/home/xxxx/pytorch-segmentation-detection/vision/torchvision/models/resnet.py in conv3x3(in_planes, out_planes, stride, dilation)
     35 
     36     return nn.Conv2d(in_planes, out_planes, kernel_size=kernel_size, stride=stride,
---> 37                      padding=full_padding, dilation=dilation, bias=False)
     38 
     39 

~/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/conv.py in __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, groups, bias)
    276         super(Conv2d, self).__init__(
    277             in_channels, out_channels, kernel_size, stride, padding, dilation,
--> 278             False, _pair(0), groups, bias)
    279 
    280     def forward(self, input):

~/anaconda3/envs/py36/lib/python3.6/site-packages/torch/nn/modules/conv.py in __init__(self, in_channels, out_channels, kernel_size, stride, padding, dilation, transposed, output_padding, groups, bias)
     31         else:
     32             self.weight = Parameter(torch.Tensor(
---> 33                 out_channels, in_channels // groups, *kernel_size))
     34         if bias:
     35             self.bias = Parameter(torch.Tensor(out_channels))

TypeError: torch.FloatTensor constructor received an invalid combination of arguments - got (int, int, numpy.int64, numpy.int64), but expected one of:
 * no arguments
 * (int ...)
      didn't match because some of the arguments have invalid types: (int, int, numpy.int64, numpy.int64)
 * (torch.FloatTensor viewed_tensor)
 * (torch.Size size)
 * (torch.FloatStorage data)
 * (Sequence data)

[warmspringwinds]

@coolcoder001 @superoil1983 @sharanry -- that's a bug that appeared after the new version of pytorch came out. Will resolve it soon. Thank you for reporting guys :)

[warmspringwinds]

@coolcoder001 @superoil1983 @sharanry

I have just checked the file resnet_34_8s_benchmark with pytorch version of 0.3.1 and python 2 installed in anaconda 2 and it seems to work fine.

Could you please provide the python version/pytorch version and python version that you guys are using when experiencing this problem?

Thank you.

[warmspringwinds]

I have also checked it on the latest unstable version of pytorch and python2 -- it works.

My hypothesis is that this problem is related to python 3 which we don't support yet. Should work on python 2 -- let me know if you have any further problems.

[vishalanand]

@warmspringwinds Could you please put in the requirements.txt file as well? Will clear out a lot of confusions.

pip freeze > requirements.txt

---

P.S. In my case, the errors are regarding unknown arguments for ResNet initialization. Tried on multiple cloud / GPU setups with same output. Maybe there are unchecked files (there could be cached files in your setups. On a clean repo pull, these errors might occur for you as well) [python2.7, pytorch-0.3.1]:

TypeErrorTraceback (most recent call last)
<ipython-input-1-a4cbdc8e5706> in <module>()
     36 
     37 print(torch.__version__)
---> 38 fcn = resnet_dilated.Resnet34_8s(num_classes=21)
     39 fcn.load_state_dict(torch.load('resnet_34_8s_68.pth'))
     40 #fcn.load_state_dict(torch.load('resnet34-333f7ec4.pth'))

/models/pytorch-segmentation-detection/pytorch_segmentation_detection/models/resnet_dilated.py in __init__(self, num_classes)
    290         # Load the pretrained weights, remove avg pool
    291         # layer and get the output stride of 8
--> 292         resnet34_8s = models.resnet34(fully_conv=True, pretrained=True, output_stride=8, remove_avg_pool_layer=True)
    293         #resnet34_8s = models.resnet34(pretrained=True)
    294 

/usr/local/lib/python2.7/dist-packages/torchvision/models/resnet.pyc in resnet34(pretrained, **kwargs)
    172         pretrained (bool): If True, returns a model pre-trained on ImageNet
    173     """
--> 174     model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
    175     if pretrained:
    176         model.load_state_dict(model_zoo.load_url(model_urls['resnet34']))

TypeError: __init__() got an unexpected keyword argument 'fully_conv'

Also, if you could also put in the resnet_34_8_66.pth, it will be helpful to execute the code without changes in resnet_34_8s_demo.ipynb

Many thanks!

[vakkov]

The way I got it to work in Python 3 was to change:

--- a/pytorch_segmentation_detection/models/resnet_dilated.py +++ b/pytorch_segmentation_detection/models/resnet_dilated.py @@ -1,6 +1,9 @@ import numpy as np import torch.nn as nn -import torchvision.models as models +import sys +#sys.path.insert(0, '/home/vakko/Downloads/disso/pytorch-segmentation-detection/vision/') +import vision.torchvision.models as models +print(models.__file__) aaand to import torchvision in the same way in the notebook; in my case - multiclass_resnet_18_8s_train:

• "import torchvision\n",
• "import vision.torchvision as torchvision\n"